LASER: Loss-Aware Singular-value Decomposition and Rank Allocation for Efficient Low-Precision Vision-Language Models

Haiyu Wang, Yutong Wang, Leshu Li, Yihui Ren, Sai Qian Zhang · 2026

Read LASER: Loss-Aware Singular-value Decomposition and Rank Allocation for Efficient Low-Precision Vision-Language Models by Haiyu Wang, Yutong Wang, Leshu Li, Yihui Ren, Sai Qian Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-00573.txt.

Description

Haiyu Wang and co-authors shrink sight-word models by a loss-wise split of their weight grids. Such models cost much to run on small gear. Low-rank splits can shrink them, but past tools chase grid fit, not task loss. Rank picks were flat or rule-of-thumb in those tools.

Their frame, LASER, weights the split by bend in the model loss. A split-form Fisher gauge steers the split toward end task marks. Rank shares cross layers by marks from tune runs. Feed-forward layers, long left out, join the shrink plan here. A mix with grade cuts serves those layers in the frame. Tests show more than twice the read speed of past work. Marks hold strong at low grade in their runs.

Catalog record

CREATORHaiyu Wang, Yutong Wang, Leshu Li, Yihui Ren, Sai Qian Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85 KB
LICENSEhttp://creativecommons.org/publicdomain/zero/1.0/
RIGHTSCC0
SOURCEarXiv

Source: https://arxiv.org/abs/2606.00573

Browse the full library · About the Tower of Babel